Gemini listens to YouTube and Reddit. ChatGPT almost never does.

1. Each assistant has its own ears

Gemini cites YouTube in 53 of its 100 answers and Reddit in 48. ChatGPT cites YouTube in 1 and Reddit in 0: it leans on test and review publications instead. Same questions, same day, two different sets of sources.

Perplexity is missing from this section on purpose. Its answers name their sources in words (“Wirecutter’s top pick”) but only 3 of 99 carried a link we could read, so we do not rank what it cites.

2. The brand’s own site is rarely the source

Of the 210 pages ChatGPT cited, 51 belong to a brand the same answer recommends. For Gemini it is 34 of 659. The rest are third parties: reviewers, comparison sites, forums, videos. An assistant recommends a brand on what others say about it.

3. ChatGPT’s answers are short lists

ChatGPT names 2.4 brands per answer on average, Gemini 5.4 and Perplexity 6. Across the 96 questions all three answered with brands, Gemini and Perplexity recommended 737 brands between them, question by question. 512 of those are not in ChatGPT’s answer to the same question.

Part of this is how we asked: one web search for ChatGPT, and a small, fast model. A longer answer from a larger model would likely name more brands. What the count shows is that the same question can return two names in one place and six in another.

4. They agree more than they differ

We expected three assistants to give three answers. They did not. In 86 of 96 questions, the brand ChatGPT puts first is also recommended by both others. All three put the same brand first in 40 questions, and three different brands first in only 9.

40 of 96

Same brand first on all three

47 of 96

Two of the three share the first brand

9 of 96

Three different brands first

Method and limits

• 100 questions in 10 sectors, written without any brand name. Each one ended with “I'm in the United States.” for all three assistants: in a first run from Spain without it, Gemini answered for a Spanish buyer and the other two for an American one.
• One answer per question and per assistant, collected on 13 November 2026 through the APIs: OpenAI’s Responses API with web search, Perplexity’s Agent API, Gemini with Google Search grounding. An answer given without a web search was asked again, then dropped.
• 299 of 300 answers came back. 96 questions have brands from all three and are the base of every comparison. In the others, one assistant failed or named no brand.
• The brands were listed from each answer by a language model, then merged when two names mean the same seller (Alienware and Dell, Max and HBO Max). We checked the lists by hand on 20 answers. A different reader would settle a few cases differently.
• A page counts as a source when the answer links it. A page belongs to a brand when the site’s name contains a brand the same answer recommends.
• The questions are in English, about products sold in the United States. Nothing here says what the same assistants answer in another language or another country.

The questions and the data

Every question, every answer and the brands read in each one are in a single file. Download the data (JSON).